Multiplier XSIX: open models you can run privately.
Multiplier XSIX is our own open-weight model family, built to finish business workflows, not just answer questions. Two tiers, Fast and Deep, built on NVIDIA Nemotron 3 and made to run on hardware you control.
Real-time agents, voice and high-volume automation. Built on NVIDIA Nemotron 3 Nano: 30B parameters with only 3B active per token, so it's fast enough for live calls and cheap to run at volume. Fits on a single data center GPU.
Long, multi-step reasoning and complex workflows. Built on NVIDIA Nemotron 3 Ultra: 550B parameters with 55B active per token. Reads whole contracts, policy libraries and audit trails in one pass. Runs on a single 8-GPU server.
Use it in the playground, through an OpenAI-compatible API, or privately on your servers, in your cloud, or in our SOC 2 compliant facility in Atlanta.
Most AI tools rent a general model and wrap it in a prompt. XSIX is trained to complete business workflows: valid tool calls, correct records, fewer steps. You get open weights you can inspect, host and keep.
XSIX Fast handles most of the traffic and passes hard steps to XSIX Deep. Both come from the same model family, so handoffs are clean, and Fast can draft tokens that Deep checks to speed Deep up.
Human approval is part of the model, not an afterthought. Approval-gate tokens pause an agent before risky actions like payments or record deletions, and a dedicated tool-call head keeps function calls in valid format.
In the playground, through an OpenAI-compatible API, or in your own environment. Run it on your servers, in your cloud, or on dedicated NVIDIA GPUs in our SOC 2 compliant facility in Atlanta. Prompts, data and outputs stay inside the boundary you choose.
Rewarded only when the workflow actually finishes correctly. Training rewards completed tasks, schema-valid tool calls, respected approval gates and fewer steps. Base and teacher models are US-origin, and client data is never pooled without opt-in.
"We built XSIX so the AI running your workflows can live on hardware you control, and finish the job instead of just answering questions." Grayson Mylar, Founder and CEO
Want the architecture and training details? Ask an engineer for the XSIX technical brief.
Multiplier XSIX is a family of two open-weight models from ALLTIPLY Labs that companies can host privately. XSIX Fast runs real-time agents, voice and high-volume automation. XSIX Deep handles long, multi-step reasoning. Both are post-trained versions of NVIDIA Nemotron 3 models, built to complete business workflows.
Can I try it before we talk?
Yes. Bring a real business question to the AI playground and compare how XSIX Fast and XSIX Deep handle it.
What are the base models and licenses?
XSIX Fast is built on NVIDIA Nemotron 3 Nano, under the NVIDIA Nemotron Open Model License. XSIX Deep is built on NVIDIA Nemotron 3 Ultra, under OpenMDW-1.1, a permissive Linux Foundation license. Both allow commercial use.
Can it run in our own environment?
Yes. XSIX Fast fits on a single data center GPU, and XSIX Deep runs on one 8-GPU server. Run either on your hardware, in your private cloud, or in our SOC 2 compliant facility, through an OpenAI-compatible API.
Is this what you use in custom builds?
When a system benefits from an owned model, yes. We can fine-tune a version of XSIX on your data, or pick a different model if it fits the job better. We don't push our own model when another one is the right call.
